Nvidia made $81.6bn in a quarter. The boom is extremely real.

The 20 May print, in full: record $81.6bn revenue, up 85% year on year and 20% sequentially. $75.2bn from data centre alone (up 92%). The rest — gaming, professional visualisation, automotive — barely moves the needle. Then an $80bn buyback top-up and a dividend bump to $0.25 a share, because when you're printing this much free cash the question becomes what to do with it. And what Nvidia has decided to do with it is hand it back rather than invest it internally. File that.

One company's data-centre line is bigger than the total revenue of almost every firm in this industry. Every model you use — from the closed-weight frontrunners to the open-weight flood — ultimately rents Nvidia silicon. That's not a moat story; it's a single-point-of-dependency story. If Nvidia's roadmap slips, if a fab has a bad quarter, or if demand digests, it doesn't dent one firm — it reprices the whole field, simultaneously. The AI and Nvidia trades are, for now, the same trade. The 'AI is real' optimists and the 'AI is a bubble' sceptics are both partly right — and they're both looking at the wrong risk. The risk isn't fakeness. The risk is concentration.

Set the numbers side by side and the picture is stark:

The 92% share going to data centre is the number to sit with. Nvidia has effectively become a pure-play AI infrastructure company. Which means any slowdown in AI infrastructure spending is a direct hit, with no other segment to absorb it.

The 85% growth headline is real but it obscures a more interesting structural shift. When data centre is 92% of your revenue, you are not a diversified technology company — you are a pure-play AI-infrastructure bet that happens to still make gaming cards on the side. That matters for how you read the $80bn buyback: Nvidia isn't diversifying into robotics, sovereign cloud, or automotive at the scale the 'largest infrastructure expansion in history' narrative would suggest. It's handing the cash back. Which is fine, and arguably correct, but it's not the posture of a company confident it can deploy that capital at internal returns better than its own share price — it's the posture of a company that has run out of ways to reinvest faster than TSMC can supply wafers. Both readings fit the facts.

Next checkpoint is Rubin, announced in January 2026 and now confirmed in production with partner availability in H2 2026. If it lands cleanly and on time, the bull case for extending the run is intact and the bears wait another year. If it stumbles — fab delays, power-delivery issues, a competitor catching up faster than expected — you'll find out very quickly how much of the AI trade was really a Nvidia trade. The answer is: almost all of it. Watch the Q2 hyperscaler capex guidance from Microsoft, Google, Amazon, and Meta. When that number turns, the digestion phase has arrived regardless of what Rubin does.

So: the boom is real, the numbers are staggering, and the right fear isn't 'is this fake?' The right fear is 'what does it look like when the whole sector is this dependent on one chipmaker's execution?' The 85% growth figure is the headline. The 92% revenue concentration in a single segment — and the $80bn buyback instead of reinvestment — are the story.